Samsung and Naver launch the latest AI chips that are 8 times more energy efficient than NVIDIA

Mondo Technology Updated on 2024-01-30

Samsung Electronics and N**ER showcased the artificial intelligence (AI) semiconductor they jointly developed in the recent year, which is about 8 times more energy efficient than competitor chips such as Nvidia and is expected to power N**ER's hyperscale AI model, Hyperclova X.

The Ministry of Science and ICT of the Republic of Korea held the "Fourth AI Semiconductor High-level Strategic Dialogue" to showcase the achievements of Korean AI semiconductor companies. The AI semiconductor developed by N**er and Samsung Electronics was unveiled in the form of a field-programmable gate array (FPGA), which is expected to be used in Hyperclova X running N**ER.

The AI semiconductors developed by N**er and Samsung Electronics are unveiled in the form of field-programmable gate arrays (FPGAs). FPGAs are a type of semiconductor that allows developers to modify designs that are primarily used for prototype production before mass production. This semiconductor is specifically used for "inference," the process by which an AI model uses new data to produce a logical result after it has completed the "learning" phase.

N**er says the AI semiconductor is eight times more powerful than competing products from companies like NVIDIA. This is achieved by integrating low-power, compact double data rate (LPDDR) DRAM to improve efficiency. Higher power efficiency means that the same computing performance can be achieved with less power consumption, enabling large-scale computing tasks to be performed at a lower cost.

Previously, the two companies had signed a memorandum of understanding (MOU) in December last year to develop AI semiconductor solutions and set up a practical working group.

South Korea** plans to strengthen domestic AI semiconductors in three phases by 2030. The first phase involves applying Korea's domestically produced neural processing units (NPUs) to data centers and providing cloud-based AI services.

Related Pages